A System for Continuous Reservoir-Simulation-Model Updating and Forecasting

Abstract

A significant body of work has demonstrated both the promise and difficulty of quantifying uncertainty in reservoir simulation forecasts. It is generally accepted that complete and reliable quantification of uncertainty should lead to better decision making and greater profitability. Many of the techniques presented in previous work attempt to quantify uncertainty without sampling the full parameter space, saving on the number of simulation runs, but inherently biasing and underestimating the uncertainty in the resulting forecasts. In addition, much of previous work has looked at uncertainty quantification in synthetic models and does not address the practical issues of quantifying uncertainty in an actual field. Both of these issues must be addressed in order to reliably quantify uncertainty in practice. In this study a new approach to reservoir simulation is proposed whereby the traditional one-time simulation study is replaced with a continuous process potentially spanning the life of the reservoir. In this process, reservoir models are generated and run 24 hours a day, seven days a week, allowing many more runs than previously possible and yielding a more thorough exploration of possible reservoir descriptions. In turn, more runs enables better estimates of uncertainty in resulting forecasts. This is combined with real-time acquisition of production and pressure data, which is automatically integrated into simulation runs to allow the process to run continuously with little human interaction. Two tests of this continuous simulation process were conducted. The first test was conducted on the Production with Uncertainty Quantification (PUNQ) synthetic reservoir. Comparison of our results with previous studies shows that the continuous approach gives consistent and reasonable estimates of uncertainty. The second test was conducted in real time on a live field. This test demonstrates the continuous simulation process and shows that it is feasible and practical for real world applications. Introduction Reservoir management is generally considered a continuous process that should span the entire life of a reservoir.1–2 Furthermore, reservoir simulation, with its unique predictive capabilities, is widely regarded as a critical tool in modern reservoir management practice.3 Reservoir simulation yields an assessment of reservoir properties and, when a forecast run is made, an assessment of future production. These assessments feed directly into the decision-making process. In his rules for decision making, Howard 4 establishes that it is necessary to assign probabilities to all possible outcomes of uncertain events. Therefore, making a good decision requires taking into account all possible outcomes and so it is necessary to quantify the uncertainty in forecasts. Conversely, if the uncertainty quantification in a forecast is incomplete, or nonexistent, then the decision may be poor. For this reason it is necessary to rigorously quantify uncertainty in production forecasts. Capen 5 demonstrated thirty years ago that people in the petroleum industry significantly underestimate uncertainty in their assessments. In keeping with this tendency, reservoir simulation engineers traditionally take only limited consideration of uncertainty and often times do not try to quantify it at all. Quantifying uncertainty in production forecasts, of course, is not a trivial undertaking. The reservoir parameter space, the set of all possible combinations of reservoir parameters, is literally infinite. Attempts at uncertainty quantification in more recent studies, specifically Floris et al.,6 have shown that, even when we explicitly try to quantify uncertainty in simulation studies, we still tend to underestimate it. It is therefore worthwhile to explore reservoir simulation techniques aimed at better quantifying uncertainty in forecasts. Typically, reservoir simulation is only utilized at discrete points in the life of a reservoir. Reservoir studies are expensive and time-consuming due to the time and manpower required to tune and history match a simulation model. As such, traditional simulation studies usually can only be justified when considering a major investment. Taken individually, smaller reservoir management decisions often do not warrant the expense of a simulation study and thus must be made without simulation results. Inaccurate forecasts or no forecasts at all can lead to sub-optimal operations and significant economic consequences. Clearly, reservoir management would benefit if a calibrated simulation model was available at any time.

A hybrid differential evolution algorithm approach towards assisted history matching and uncertainty quantification for reservoir models

“…This will become a limitation for HDE when using parallel computing where number of iterations is more restrictive than number of simulations. HDE gave equally good results even when the search started from (Floris et al, 2001;Demyanov et al, 2004;Holmes et al, 2007;Hajizadeh et al, 2010); HDE-NAB (this work). different models.…”

“…Fig. 14 shows a comparison of these approaches with this work of the same PUNQ-S3 model (Floris et al, 2001;Demyanov et al, 2004;Holmes et al, 2007;Hajizadeh et al, 2010). With this study, it can be inferred that the approaches which show large deviation of FOPT prediction with the truth case with small uncertainty range may lack proper exploration of the search space.…”

“…The PUNQ-S3 is a synthetic benchmark model which is used for comparing the performance of different history matching and uncertainty quantification techniques (PUNQ-S3, 2015). The same has been used by various researchers for history matching and uncertainty quantification studies (Floris et al, 2001;Demyanov et al, 2004;Holmes et al, 2007;Hajizadeh et al, 2010).…”

Field Development Process Revealing Uncertainty Assessment Pitfalls

“…For HM, although various techniques may apply (Holmes et al, 2007;Liu e McVay, 2010; Oliver e Chen, 2011), this paper uses an ensemble-based method (Aanonsen et al, 2009;Emerick, A. A., & Reynolds, 2013) as we need to history match a large number of simulation models.…”